Multimodal AI
Why CCTV Alone Cannot Explain Retail Conversion
CCTV alone cannot explain retail conversion because a camera measures the visit and conversion is decided in the conversation. A camera can count that 300 people entered and 110 billed. It can show that 40 of the 190 who left were never approached. It cannot say what the other 150 asked for, what they were quoted, what they objected to, or whether anyone asked for their number. This guide walks through what CCTV conversion tracking actually computes, where it misleads, and what the conversation adds.
What CCTV conversion tracking computes
The standard calculation is bills divided by entries, sometimes by hour and sometimes by associate. Better systems exclude staff, count groups as one buying unit, and add dwell by zone and wait before engagement. All of this is useful and all of it describes the visit from the outside. It is the record of what happened around the interaction, not inside it. What happened inside, the intent, the objection, the offer and the outcome, lives in the conversation, and the conversation is not on the camera.
Six things the camera sees but cannot read
- A ten-minute conversation at the desk. Interest, or a slow advisor, or a finance explanation that went wrong.
- A customer who leaves after the offer. Price too high, wrong colour, or a spouse to consult tonight and a return tomorrow.
- A group of three. Which one decides, and whether the advisor spoke to them.
- A long dwell at the premium display. Desire, or confusion about the price tag.
- A customer who leaves with a smile. Happy, or polite. Neither is a sale.
- Two customers with identical paths, one billed and one did not. The difference was spoken.
One visit, as the camera saw it and as it was said
| Time | What the camera recorded | What the consented conversation recorded |
|---|---|---|
| 6.02 pm | Solo entry, male, mobile phone store, Lucknow | (no conversation yet) |
| 6.02 to 6.05 | Three minutes at the mid-range display wall | (no conversation yet) |
| 6.05 | Advisor arrives; both move to the counter | Advisor taps to record; customer agrees. Asks for a specific model, 8 GB, under 20,000 |
| 6.05 to 6.11 | Six minutes at the counter; customer handles two devices | Advisor shows the asked-for model and one above it. Quotes the exchange offer at the right amount. Customer says a rival chain quoted 1,500 less on the same model |
| 6.11 to 6.13 | Customer looks at phone; advisor steps away briefly | Advisor checks with the manager on a price match. Customer says he will wait |
| 6.13 | Advisor returns; short exchange | Match refused; advisor offers a free case and screen guard instead. Customer says he will think about it. Advisor asks for a number; customer gives it and says call by Thursday |
| 6.14 | Customer exits without passing billing | Conversation ends. Lead created: model, budget, rival quote, callback by Thursday |
Where camera data misleads on its own
The visit above is an unconverted twelve-minute solo visit with a three-minute wait, which is what the camera report will say. On that reading the store lost a customer at the counter after a long interaction, and a regional manager might ask the advisor what took so long. The conversation says the opposite: the advisor ran the playbook, quoted the offer right, met a rival quote he could not match, and captured a lead with a date. The store has a pricing problem with one rival on one model, and a callback to make by Thursday. Camera data was not wrong. It was silent on everything that mattered.
The same misreading happens in aggregate. Long dwells are read as engagement when they are often waiting. Short visits are read as failures when some are quick, correct answers to a customer who came for one thing you did not stock. Group visits are counted as one buyer, and the camera cannot see that the buyer was the person who said nothing.
What the camera is still for
None of this argues against the camera. It is the only instrument that sees the customers who never spoke to anyone, and unattended walk-outs are the single biggest conversion leak in many Indian stores. It supplies the denominator that keeps every conversation score honest: consented conversations over entries. It sees the wait, the queue at billing and the group. What it needs is the other signal beside it.
Adding the conversation to the camera
- Keep the camera events: entries, zone dwell, wait before engagement, group size, walk-outs. Process them on an edge device in the store so frames never leave.
- Add consented capture on the advisor's phone, started with a tap after the customer agrees.
- Join the two on that tap, by time, store and zone. No identity is needed, and none is used.
- Score each join. One track at the counter at tap time is strong; a crowded counter with two advisors is weak, and it is reported that way.
- Read the store-week: conversion by wait, objections by hour, unattended walk-outs next to capture rate.
- Borentis runs the conversation half today, as Borentis Floor, in Hindi, English and Hinglish. ShopperDNA, which joins camera events to those conversations, is on the roadmap.
Frequently asked questions
Can CCTV analytics tell me why customers leave without buying?
No. It can tell you how many left, how long they stayed, whether they were approached and where they stood. The reason was spoken, and the camera did not hear it.
Can we use the audio from our CCTV instead?
No. CCTV audio is unconsented, poor quality and legally weak under India's DPDP Act. The conversation must be captured with consent, which in practice means on the advisor's phone after the customer agrees.
Is per-associate conversion from CCTV a fair measure?
It is a fair count of outcomes and an unfair judgement of cause. The visit above would count against a good advisor. Adherence by step from the transcript is the fair measure; camera conversion is its context.
Related reading
- How to use CCTV for retail analytics
- Why retailers need conversation intelligence and computer vision together
- Conversation intelligence vs CCTV and footfall analytics
- ShopperDNA, on the roadmap
Where Borentis applies this
- Walk-in Recovery: The customer who left is still yours.
- Objection Intelligence: The reason they did not buy, in their own words.
- Execution Scorecards: See the floor before the P&L does.
Borentis is the Agentic Operating System for Customer Interactions, built for Indian retail floors: consented one-tap capture on the advisor's phone, every conversation scored against your playbook with the evidence behind every number, leads created when a number is heard, and coaching from your own best conversations.